A New Fitness Function for Evolutionary Music Composition

Abstract

In this paper we propose a new fitness function for Evolutionary Computation purposes, based on a weighted by neighborhood average distance between two sequences of points within any metric space. We will apply this fitness function to the field of Computer-Assisted Composition focusing on the problem of thematic bridging, consisting in the evolutionary creation of a soft set of transitions between two given different melodies, the initial and the final one. Several self-adaptive strategies will be used to perform the search. A symbolic melody will be genotypically mapped into a sequence of genes, each of then containing the information of duration, frequency and time distance to following note. We will test the implementation of the fitness function by means of two experiments, showing some of the intermediate melodies generated in a successful run, and benchmarking every experiment with performance indicators for any of the three distinct evolutionary strategies implemented. The results prove this novel fitness function to be a quick and suitable way for individual evaluation in genetic algorithms.